Model comparison
Kimi K2 (Jul 2025) vs Kimi K3
Kimi K3 is the stronger model overall, scoring 59.5 to 41.2 on the Noometry Index. Kimi K2 (Jul 2025) costs 6.0× less per token, which makes it the better buy when Kimi K3's lead doesn't matter for your workload.
Last verified . 23 shared benchmarks.
Summary
- They share 23 benchmarks with published results for both. Kimi K2 (Jul 2025) scores higher in 0 categories and Kimi K3 in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Kimi K3 leads 63.0 to 23.3.
- The biggest single-benchmark swing is WeirdML: 42.8% for Kimi K2 (Jul 2025) and 82.6% for Kimi K3.
- Kimi K2 (Jul 2025) is cheaper at $0.57 / $2.30 per million input/output tokens, against $3 / $15 for Kimi K3.
- Kimi K3 accepts more context: 1.05M tokens versus 262K.
Side by side
| Kimi K2 (Jul 2025) | Kimi K3 | |
|---|---|---|
| Provider | Moonshot AI | Moonshot AI |
| Noometry Index | 41.2 | 59.5 |
| Released | 2025-07-12 | 2026-07-16 |
| Weights | Open | Open |
| Context window | 262K | 1.05M |
| Max output | 262K | 1.05M |
| Input $ / M tokens | $0.57 | $3 |
| Output $ / M tokens | $2.30 | $15 |
| Results tracked | 42 | 53 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Kimi K3 leads
Kimi K2 (Jul 2025): 42.4 (#102), Kimi K3: 61.0 (#10)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| WeirdML | 42.8% | 82.6% |
| LMArena Coding | 1399 | 1508 |
| ALE-Bench | 597.5 | 1,524 |
| DeepSWE | — | 68.5% |
| FrontierCode | — | 44.2% |
| SWE-bench Verified (bash only) | 63.4% | — |
| Aider Polyglot | 59.1% | — |
| LMArena WebDev | — | 1654 |
| FrontierSWE | — | 25.9% |
| SciCode | — | 59.5% |
| GSO | 4.9% | — |
Agentic & Tool Use Kimi K3 leads
Kimi K2 (Jul 2025): 32.4 (#64), Kimi K3: 41.8 (#20)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| Terminal-Bench | 35.7% | — |
| APEX-Agents | — | 50.6% |
| Berkeley Function Calling Leaderboard | 59.1% | — |
| τ²-bench Banking | — | 37.1% |
| PostTrainBench | — | 32% |
| GBAEval | — | 48.3% |
| GDP.pdf | — | 19% |
| METR Time Horizons | 59.2% | — |
| Vending-Bench 2 | — | 5,165 |
Reasoning Kimi K3 leads
Kimi K2 (Jul 2025): 23.3 (#179), Kimi K3: 63.0 (#17)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| SimpleBench | 26.3% | 60.7% |
| LMArena Hard Prompts | 1384 | 1496 |
| Epoch Capabilities Index | 146.01 | 157.45 |
| ForecastBench | 60.2 | 61.1 |
| ARC-AGI-2 | — | 60.4% |
| Kagi LLM Benchmark | 64.4% | — |
| NYT Connections (extended) | — | 93.6% |
| ARC-AGI-1 | — | 94.5% |
| CritPt | — | 23.4% |
| Chess Puzzles | — | 39% |
| Mystery Game Puzzles | — | 26% |
| DTBench | — | 91.2% |
| LMCA | — | 52.7% |
| Surface Evolver Bench | — | 95% |
Math Kimi K3 leads
Kimi K2 (Jul 2025): 42.7 (#83), Kimi K3: 74.2 (#16)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Math | 1397 | 1491 |
| FrontierMath (Tiers 1-3) | — | 72.2% |
| FrontierMath Tier 4 | — | 39% |
| MathArena Final-Answer Competitions | — | 87.8% |
| OTIS Mock AIME 2024-2025 | — | 97.2% |
| ProofBench | — | 87% |
| Omni-MATH | 65.4% | — |
| FrontierMath (Feb 2025 set) | 21.4% | — |
| FrontierMath Tier 4 (v1) | 0% | — |
Knowledge Kimi K3 leads
Kimi K2 (Jul 2025): 37.3 (#157), Kimi K3: 63.2 (#21)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Expert | 1365 | 1521 |
| GPQA Diamond | — | 93.1% |
| SimpleQA Verified | — | 50.6% |
| MMLU-Pro | 81.9% | — |
| Confabulations | 20.4% | — |
| Vectara Hallucination Rate | 17.9% | — |
| GPQA (HELM) | 65.3% | — |
Multimodal Not comparable
Kimi K2 (Jul 2025): —, Kimi K3: 37.8 (#70)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| Blueprint-Bench 2 | — | 29.5% |
| Furniture Assembly | — | 34.2% |
Multilingual Kimi K3 leads
Kimi K2 (Jul 2025): 49.6 (#130), Kimi K3: 56.3 (#21)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Non-English | 1372 | 1466 |
| LMArena Chinese | 1415 | 1529 |
| LMArena French | 1379 | 1491 |
| LMArena German | 1387 | 1488 |
| LMArena Japanese | 1349 | 1487 |
| LMArena Korean | 1325 | 1458 |
| LMArena Russian | 1385 | 1482 |
| LMArena Spanish | 1386 | 1472 |
Instruction Following Kimi K3 leads
Kimi K2 (Jul 2025): 71.1 (#156), Kimi K3: 77.7 (#14)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Instruction Following | 1348 | 1483 |
| IFEval | 85% | — |
Long Context Kimi K3 leads
Kimi K2 (Jul 2025): 41.2 (#145), Kimi K3: 45.8 (#29)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Longer Query | 1353 | 1494 |
| Fiction.LiveBench | 66.7% | — |
| CL-bench | 17.6% | — |
Writing & Preference Kimi K3 leads
Kimi K2 (Jul 2025): 62.3 (#78), Kimi K3: 76.6 (#4)
| Benchmark | Kimi K2 (Jul 2025) | Kimi K3 |
|---|---|---|
| LMArena Text | 1380 | 1476 |
| LMArena Creative Writing | 1350 | 1454 |
| EQ-Bench Creative Writing | 1666 | 2082 |
| LMArena Multi-Turn | 1371 | 1488 |
| Short-Story Creative Writing | 85.6% | — |
| WildBench | 86.2% | — |
| EQ-Bench 4 | — | 1339 |
Frequently asked questions
Is Kimi K2 (Jul 2025) better than Kimi K3?
Kimi K3 is the stronger model overall, scoring 59.5 to 41.2 on the Noometry Index. Kimi K2 (Jul 2025) costs 6.0× less per token, which makes it the better buy when Kimi K3's lead doesn't matter for your workload.
Which is cheaper, Kimi K2 (Jul 2025) or Kimi K3?
Kimi K2 (Jul 2025) is cheaper. It lists at $0.57 per million input tokens and $2.30 per million output tokens; Kimi K3 lists at $3 and $15.
Is Kimi K2 (Jul 2025) or Kimi K3 better for coding?
Kimi K3 scores higher on coding benchmarks: 61.0 versus 42.4 in the Noometry coding category.
Which has the bigger context window?
Kimi K3 does, with 1.05M tokens against 262K.
How many benchmarks do Kimi K2 (Jul 2025) and Kimi K3 share?
23 benchmarks have published results for both models. Kimi K2 (Jul 2025) has 42 scored results on Noometry and Kimi K3 has 53.